Genetic determinants of growth hormone and GH-related phenotypes

Growth hormone (GH) is the main regulator of longitudinal growth during childhood and adolescence. In addition, this versatile hormone has effects on lipolysis in adipose tissue, protein anabolism in muscle cells and glucose and lipid metabolism throughout the entire life [1]. We recently showed that high fasting levels of growth hormone are associated with cardiovascular disease and that higher levels predict cardiovascular morbidity and mortality independently of traditional cardiovascular risk factors confirming previous reports from the 1990’s [2, 3]. However, whether or not this association is causal is unknown. Causality can be assessed by randomized controlled trials (RCTs), administering GH or blocking its effect, and test if such treatment affects the risk of cardiovascular disease. However, such studies are unlikely to take place due to large costs. In light of this, Mendelian Randomization studies have established themselves as a cost effective alternative to RCTs as they have convincingly been shown to be able to assess causality between circulating biomarkers and medical traits and diseases [4, 5]. In order to enable tests of causality between GH and cardiometabolic diseases using Mendelian Randomization, we here set out to identify common genetic variants contributing to fasting GH concentration.

GH is secreted in pulses, which makes measurement difficult. However, the levels of GH are quite low and stable during the morning hours [2, 6–8], which enable measurement of fasting GH concentrations. Even if the variability in single values of fasting GH is large, there is a strong positive association to 24-h GH production on a group level [9].

The aim of this study was to find genetic determinants of the fasting level of growth hormone. We applied two different strategies, the first being a GWAS-approach where we related single nucleotide polymorphisms (SNPs) from the whole genome to hs-GH in the Cardiovascular Cohort of the Malmö Diet and Cancer study (MDC-CC). The top SNPs were then replicated in another non-overlapping cohort, the Malmö Preventive Project (MPP). The final step would be to genotype the best performing SNP in more than 24,000 people in the whole Malmö Diet and Cancer study (MDC) and relate it to known GH-associated variables such as height and BMI. In the second approach, which was a candidate gene approach, we analyzed SNPs in the growth hormone gene (GH1) [10] and the growth hormone receptor gene (GHR) [11] in the MDC-CC with subsequent replication in the MPP and MDC as previously described.

Baseline characteristics of the three different cohorts are shown in Table 1. There were more males in MPP compared with the other cohorts. The participants in MPP were also older, had a larger waist and were more likely to be on lipid lowering therapy. The differences in hs-GH between males and females are less pronounced in MPP when compared with MDC-CC. As stated in the methods section, 53% of the MPP-cohort participated in MDC and was excluded before the random selection for the hs-GH and genetic analyses, baseline characteristics of these are available in Additional file 1: Table S1.

Candidate gene approach

We also performed a candidate gene approach in our search for genetic determinants of the fasting level of hs-GH. SNPs in the genes GH1 and GHR were genotyped and then related against the fasting level of hs-GH. In our assay there were 2 SNPs in the gene GH1, however both of these had a MAF < 0.01 which was our limit for further analysis. In the gene GHR there were 49 SNPs, but in 17 of these the MAF was below 0.01. Thus we could analyze 32 SNPs vs hs-GH. Among the whole cohort, 8 SNPs were nominally significant (p < 0.05). After stratification by sex, 9 SNPs were nominally significant in males and none were significant in females. Six of the nominally significant SNPs in all and male overlapped. After LD-analysis, four of the SNPs with independent signals in all were selected for replication in MPP (Table 4). The top SNP was rs13153388, where the minor allele was associated with increased hs-GH in the male (Beta [SD-increase of log-GH per 1 minor allele], 0.116; 95%CI, 0.029 to 0.204;p = 0.009) and gender-combined (Beta, 0.088; 95%CI, 0.029 to 0.147; p = 0.003) analyses, but not among females (Beta, 0.066; 95%CI, −0.014 to 0.145; p = 0.11). Three of the four selected SNPs were nominally significant in males as well (Additional file 1: Table S5). Of note is that the SNP rs6873545, which is a marker for GHRd3 [20], was not associated with the fasting levels of hs-GH in neither combined (p = 0.95) nor gender-stratified analyses (males, p = 0.80; females, p = 0.85).

SNP	Basepaira	Major allele	Minor allele	MAF	Beta	95%CI	P
rs13188386	42,473,555	G	A	0.26	0.085	0.018 to 0.151	0.01
rs13153388	42,507,333	T	G	0.44	0.088	0.029 to 0.147	0.003
rs4365846	42,547,255	A	G	0.24	0.087	0.018 to 0.156	0.01
rs6883523	42,564,748	G	T	0.18	0.079	0.003 to 0.155	0.04

Minor allele modeled as effect allele.The β coefficients are expressed as the increment of standardized values of the natural logarithm of hs-GH per 1 minor allele. Adjusted for age and sex

Abbreviations: MAF Minor allele frequency.Number of individuals available for analysis: 4126–4131

^aBasepair number for polymorphism (GRCh37). All the SNPs are located in the GHR-gene on chromosome 5

The four selected SNPs were then genotyped in MPP in a replication effort. However none of the candidate gene SNPs were significant in linear regression models vs hs-GH. Including both combined and gender-stratified analyses, the lowest p-value was 0.18 in the gender-combined analysis of rs13188386.

The top SNP from MDC-CC, rs13153388, was genotyped in MDC to relate it to anthropometric variables in multivariate linear regression models adjusted for sex and age (Table 3). The GH-increasing minor allele of rs13153388 was significantly associated with lower height in all (p = 0.003), and females (p = 0.003), but not in males (p = 0.26). Females were approximately 0,2 cm shorter per 1 minor allele (Fig. 2). It is noteworthy that the association with height is negative in a SNP that is associated with a higher level of hs-GH, which will be further discussed in the next section.

This is the first genome-wide analysis on the fasting level of GH that has been conducted, and the largest genetic study on the fasting level of GH in all categories. A GWAS-approach with a discovery phase in 4134 individuals and subsequent replication in a non-overlapping cohort of 5262 individuals identified one SNP on chromosome 17 in which the minor allele was associated with a reduced fasting level of hs-GH. A candidate gene approach identified SNPs in the GHR associated with the fasting level of hs-GH, which in a later phase showed strong associations to height. The main clinical implication of our findings is that genetic signals, such as those identified in the current study, are necessary to be able to assess whether the previously reported associations between high circulating GH levels in the fasted state and risk of cardiovascular morbidity and mortality are causal or not in Mendelian Randomization studies. Also, the novel locus associated with the fasting level of GH represents a completely new biological mechanism behind control of plasma GH level that deserves further study.

The minor allele of rs7208736 was associated with a significant reduction of the fasting level of hs-GH both in the discovery and replication cohort. The results were stronger in females in MDC-CC, but stronger in males in MPP. It should be noted that there were approximately half as many women in MPP as in MDC-CC, which may explain part of the reduced performance among females in MPP. Although the minor allele was associated with a reduction of hs-GH, it was also associated with a lower BMI and waist in the male part of MDC, which may seem counter intuitive since these variables are inversely related to the fasting level of hs-GH [2, 18]. An augmenting modification of the effect of hs-GH could explain the results, since this would lead to larger effects on target tissues and reduced secretion due to enhanced feedback signaling, i.e. the opposite of GH-insensitivity syndrome [21, 22]. It should however be noted that the associations between rs7208736 and BMI and waist are rather weak for such a big cohort, which underlines the need for replication of our results, especially concerning the metabolic effects of this SNP.

Rs7208736 is located in a non-coding sequence of the short arm of chromosome 17 (17p13.3). The GH-gene cluster is located on the same chromosome, but on the other arm, so any direct interaction with this gene is unlikely. However, numerous genes, which the SNP might affect through enhancers or regulatory motifs, are located nearby. Inactivation of the region around rs7208736 has been linked to several malignancies, such as lung, breast, liver, colon, kidney, and brain, with the genes HIC1 (Hypermethylated in cancer 1) and OVCA2 (Ovarian cancer-associated gene 2) being potential tumor suppressors [23, 24]. HIC1 is located approximately 150 kbp downstream of rs7208736, and could be the link between rs7208736 and GH. Among other effects, HIC1 represses transcription of SIRT1 (Sirtuin 1) [23, 25]. SIRT1 is a deacetylase that is activated by caloric restriction and in vivo in rats have been shown to negatively regulate GH-dependent IGF-1 production in the liver by deacetylation of STAT5 [26]. STAT5 is a downstream target of GHR-signaling via JAK2-activation [11]. Thus higher levels of HIC1 would inhibit transcription of SIRT1, which could augment the effect of GHR-signaling and would explain the picture with the minor allele in rs7208736 reducing levels of hs-GH but at the same time also being associated with lower BMI and waist. However this is very speculative, but nevertheless provides a plausible mechanism with support in the literature. The associations are intriguing and further studies will be needed to replicate this finding and determine the underlying pathways.

When investigated in Cardiogram the major allele of rs7208736, which is associated with a higher level of hs-GH, was associated with an OR of 1.05 per major allele for incident coronary artery disease. This is in line with previously published results of higher levels of hs-GH being associated to cardiovascular disease [2, 3]. However one could argue that it does not fit with the suggested theory in the previous section, i.e. the minor allele being associated with augmented sensitivity in the GHR-signaling. It should be noted that Cardiogram included more individuals than MDC, which may suggest that our proposed mechanism in with the minor allele of rs7208736 augmenting GH-signaling may be erroneous. Alternately you could interpret the results as reduced GH-signalling being associated with CVD, and our previous findings being explained as a compensatory mechanism [2]. This emphasizes the need for replication and further studies on the subject.

In the candidate gene approach the minor allele in rs13153388 was associated with a higher hs-GH and lower height. Height is as expected positively associated with hs-GH [18]. If the SNP is associated with a reduced function of the GHR, the consequence could be a mild form of GH-insensitivity syndrome [21, 22], with a reduced effect of hs-GH, i.e. short stature, but, due to reduced feedback signaling, still high levels of hs-GH in plasma. The effect of the SNP on adult height is small, but evident (Fig. 2).

Unfortunately, the results in the GHR-gene did not replicate in MPP. This may be attributable to differences in the composition of the cohorts. When comparing the baseline characteristics between MPP and MDC-CC, the population in MPP is older, consists of more males and a larger proportion are on medications that potentially could affect GH-levels. The fasting levels of hs-GH display a more pronounced gender difference in MDC-CC than in MPP. Notably the reduced difference in MPP is attributable to both an increase in males and a decrease in females. The gender difference in the fasting levels has previously been described to decline with age [27]. Among females, menopausal status and hormone replacement therapy might also affect the decline in hs-GH, since estrogen is positively associated with hs-GH [28]. It may be that these confounding effects have diluted the differences in GH-levels and made associations harder to capture. However there is indication of more long-term effects of GH in the anthropometric analyses with the strong connection to height, which is heritable to about 80% [29], implying that the SNP might affect the levels of GH earlier in life.

The study has some limitations. There was a fair amount of non-responders to the invitations, which may have biased selection of a more healthy population. Also, in MPP, roughly half the participants were excluded before the random selection, since they participated in the MDC (Additional file 1: Table S1). This was necessary to obtain independent cohorts, but also may have biased MPP-cohort, since willingness to participate in studies (i.e. likelihood to participate in both studies) infers a positive health selection, which could reduce the healthiest part of the cohort in the current study. The pulsatile secretion of hs-GH make measurements more difficult, but as previously discussed we have tried to eliminate this issue by drawing blood in the morning hours when the levels are more stable [2, 6–8]. It is inevitable that samples are taken from some individuals during a peak, but given the large number of participants it is still possible to analyze hs-GH on a group level. A single fasting value of hs-GH is not a standard clinical test. The correlation between fasting GH and 24-h GH secretion is strong, but with large variability [9]. However, on a population basis, a single fasting value of hs-GH strongly and independently predicts cardiovascular disease and death as we have previously shown [2], which calls for further research on the subject.

As discussed previously there were some differences in the characteristics between the MDC-CC and MPP. However this should be an advantage to evaluate if the results are applicable to a broader population. A concern could be that the individuals in MPP were old and that the function of their GH-IGF-1-axis may have been reduced.

In the GWAS-approach we did not reach genome-wide significance in our analysis, which means that the results have to be interpreted cautiously. However we managed to replicate our findings in a separate cohort, which increases the credibility of the results. This underlines the need of further replication of our results in independent cohorts. Since the genetic variant is not confirmed it is difficult to assess any causality of GH in cardiovascular disease through a Mendelian randomization.

In the candidate gene approach 32 SNPs were included in the initial analysis, a bonferroni-corrected p-value would be P = 0.002 (0.05/32), which the SNP rs13153388 misses by a fine margin. This emphasizes the need for further replication.

Both cohorts (MDC and MPP) are from the same geographic area with the same genetic background and allele frequencies of the studied SNPs were similar in all cohorts. This makes the results difficult to extrapolate to other ethnic groups. The results will need to be replicated in larger studies, as well as in populations with another genetic background to attain a larger generalizability.

In conclusion, we discovered one SNP (rs7208736) on chromosome 17 where presence of the minor allele is associated with a reduced fasting level of hs-GH. In the GHR we found one variant associated with a higher fasting level of hs-GH and lower height. Our findings need further replication in independent cohorts. If true they point at novel biological mechanisms underlying control of fasting plasma GH level and will facilitate future Mendelian Randomization efforts of investigating whether the previously described relationship between GH and cardiovascular morbidity and mortality is causal or not.